Messages in this thread | | | Date | Tue, 16 Feb 2010 08:55:46 -0600 (CST) | From | Christoph Lameter <> | Subject | Re: [PATCH 05/12] Memory compaction core |
| |
On Tue, 16 Feb 2010, Mel Gorman wrote:
> Because how do I tell in advance that the data I am migrating from DMA can > be safely relocated to the NORMAL zone? We don't save GFP flags. Granted, > for DMA, that will not matter as pages that must be in DMA will also not by > migratable. However, buffer pages should not get relocated to HIGHMEM for > example which is more likely to happen. It could be special cased but > I'm not aware of ZONE_DMA-related pressure problems that would make this > worthwhile and if so, it should be handled as a separate patch series.
Oh there are numerous ZONE_DMA pressure issues if you have ancient / screwed up hardware that can only operate on DMA or DMA32 memory.
Moving page cache pages out of the DMA zone would be good. A write request will cause the page to bounce back to the DMA zone if the device requires the page there.
But I also think that the patchset should be as simple as possible so that it can be merged soon.
> Ah, it was 2009 when I last kicked this around heavily :) I'll update > it.
But it was authored in 2009. May be important if patent or other copyright claims arise. 2009-2010?
| |